Image enhancement
Editor’s note: Robert Batemarco is vice president, director of analytics, and Cynthia Ziment Schlegel is managing director, at Ziment Associates, a New York-based research firm. Jane Dennison-Bauer is director at US WEST Dex, a Denver-based yellow pages publisher.
Since telephone directories first started offering to print ads in color, the chief benefit of color ads was considered their ability to catch the reader’s eye. Over time, more and more ads have come to be displayed in color. Clearly, when an ad is the only one on the page in color, it will stand out. However, at the other extreme, when all ads on a page are in color, none of them will stand out by dint of color alone. Thus, as the penetration of color in directory ads increases, the ability of color to catch the reader’s eye becomes a less and less viable selling point, limiting directory publishers’ ability to sell color ads, which are more profitable than black and yellow ads.
To justify to advertisers the higher price they will pay for color ads, one must show them that increased color penetration does not reduce the differential impact of color on customer response to those ads. Once higher levels of color penetration have been reached, perhaps in a few years (a random sample of a current Manhattan directory shows 29 percent penetration), successful marketing of color ads will require additional rationales.
Faced with this situation, U S WEST Dex, a Denver-based yellow page publisher, sought to reposition the marketing of its color ads by touting the ability of color to communicate key messages more effectively than standard black and yellow. To confirm that such an approach was realistic, it turned to Ziment Associates, a New York research firm.
Precise quantification
Proving that color ads are more effective than black and yellow ads, even at higher levels of color penetration, requires precise quantification of the importance of color on response to those ads. This requires isolating the impact of color from other factors influencing readers’ response to particular ads. Levels of ad color, color penetration on the page, type of product or service advertised and specific copy of the ads themselves must all be taken into account.
A commonly used technique for this kind of a problem is conjoint analysis. However, the complexities of this problem render conjoint inappropriate for a number of reasons. First, conjoint designs do not allow for differences in competitive products, which prevents them from handling the differences in color penetration among competitive ads which share the page, one of the main variables we need to examine. Second, conjoint requires scales in order to permit calculation of utilities. The decision to select an ad to call from a directory is an either/or choice, which is better modeled dichotomously than by scales. In addition, because the color level of specific ads was not independent of color penetration of the pages on which they appeared, the problem contains interactions which conjoint is unable to handle.
Finally, the conjoint design would result in too many combinations - 288, to be exact. Even with an incomplete factorial design, each respondent would have to look at 32 different stimuli. By generating respondent overload, this type of design would jeopardize the quality of the data ultimately collected.
It is important to keep the respondent interested throughout the interview. This can be accomplished by presenting the respondent’s task as concretely as possible and by keeping manageable the number of choices the respondent must make. At the same time, modeling the impact of color penetration properly requires testing at least three levels. Otherwise, we would have no way of knowing if the relationship between color penetration and ad selection is linear. (By linear we mean that the rate of ad selection changes at a constant rate for any change in the level of penetration - as opposed to the impact changing at different rates over different ranges of penetration levels.) Assuming linearity when it is not present impairs our ability to interpolate the effect of intermediate levels of penetration on ad selection.
A multidimensional solution
The solution to these research problems was three-fold. It consisted of an original experimental design, real-world props and use of a logit model.
A. Experimental design
The experimental design chosen was a full factorial design with incomplete blocs (i.e., respondents were only shown a fraction of the possible concepts). With that design, 36 concepts were tested. For each of three types of advertisers (florists, locksmiths and dentists), we tested three levels of color penetration (25 percent - i.e., two color ads out of eight on a page, 50 percent and 75 percent) and four levels of ad color (process photo, multiple color with graphical picture, knockout and yellow & black). For each of these concepts, two of the ads were test ads.
Since making selections among ads 36 different times would have proven beyond the tolerance of most respondents, each respondent actually was questioned about 12 of the concepts.
As we noted previously, three levels of penetration were offered to determine if the relationship is linear or if indeed there are diminishing returns to color ads as penetration increases beyond a certain point.
B. Real-world stimulus materials
To solve the problem of respondent boredom, the decision task they were asked to complete was made as real-world as possible. We did this by using simulacra of actual directory pages, printed by R.R. Donnelley & Sons Company on the same type of paper used in directories. (R.R. Donnelley & Sons Company is one of the two companies that prints the yellow pages for U S WEST Dex.) Each page included eight 2-3/8" x 6" ads plus two columns of additional listings. Respondents were then asked which of the eight ads they would be most likely, second-most likely and third-most likely to call. They performed this task 12 times on 12 different pages. After doing this, respondents then were shown nine of the ads separately (outside the context of the page) and asked to rank those individual ads overall and to rate each one on a battery of seven image attributes.
The use of these props made a straight phone interview impossible. Between the alternatives of mall intercept interviewing and phone-mail-phone, the latter was chosen since it is able to accommodate probability sampling. Thus, it was projectable to U S WEST Dex’s 14-state territory. In phone-mail-phone, the respondent is contacted by phone and the first part of the interview is completed. At that point respondents are asked if they are willing to continue the interview at a later date. If they agree, they are sent a package of materials by overnight courier, which they refer in completing the interview.
C. Logit model
Given the wealth of data collected in this study, several findings were arrived at through basic cross-tabular analysis. However, use of a logit model is what made this study unique and permitted it to determine if consumers are more likely to select products advertised in color. The logit model (also known as logistical regression) is a variant of multiple regression able to deal with non-metric variables and possessing certain features of discriminant analysis. It enables one to predict the value of a dependent variable based on various levels of a number of independent variables.
In the model for this study, the dependent variable was the color level (process photo, color, knockout and yellow & black) of the ad selected first. The independent variables were the heading (type of business advertised - florist, locksmith or dentist) and the degree of color penetration (25 percent, 50 percent or 75 percent) and the specific ad (two for each heading). The position of the ads on the page was held constant across all pages.
The regression coefficients (beta weights) derived from this model quantify how much better different levels of color perform. Headings and specific ads are non-metric data, so that their only allowable levels are on or off. Because penetration has a numeric value, even though respondents were only shown levels of 25 percent, 50 percent and 75 percent, the model will yield coefficients for that variable which let us interpolate how they would behave at intermediate levels.
IV. Results
Results were based on 600 interviews, in each of which 12 of the 36 total directory test pages were seen. Several important findings emerged from this study. The most important finding is that color increases ad selection even in the presence of high levels of color penetration. This can be seen in Figure 3. It shows the share of the times ads of each level of color were chosen first. For instance, when color ads (either multiple color or process photo) for florists make up 75 percent of all ads on a page, those color ads are chosen first 86 percent (44 percent for process photo and 42 percent for multiple color) of the time. Regardless of the level of color penetration or heading, color ads are selected first at least 60 percent of the time.
Of course, it is no great accomplishment for more color ads to be selected first when more of the ads on the page are in color. However, the other major finding yielded by this model is that color ads are selected first in greater numbers than the proportion of color ads on the page would lead one to expect. The extent to which the first-place selection of color exceeds color penetration does, of course, decline as color penetration increases. Figure 4 makes this clear. For instance, when only 25 percent of ads on the page were in color, 67 percent of the ads chosen first were in color, a 42 percent difference. At 75 percent penetration, 83 percent of ads chosen first were in color. This difference, although smaller in absolute terms, was still statistically significant at the 95 percent confidence level.
As important as it is to know that more customers patronize advertisers who buy color directory ads, positioning those ads more effectively requires the knowledge of why they do so as well. We were able to answer the latter question by correlating image attributes that play a major role in consumers’ decisions to respond to an ad with the color level of the ad.
Respondents were asked about the following seven image attributes of the companies whose advertisements they saw:
1. It is family oriented.
2. It is high quality.
3. It is trustworthy.
4. It has been in business for a long time.
5. It lets me know what to expect from the business.
6. It lets me know what to expect in terms of price.
7. It lets me know what to expect in terms of products and/or services.
For this exercise, respondents were shown nine of the ads they had seen in the full-page spreads. The nine they saw were rotated on a respondent-by-respondent basis from 24 test ads in total (four color levels of each of the six test ads). In this round they saw the ads individually without any surrounding ads. They were then asked to rank the ads in order of their likelihood of calling that advertiser. They were then asked to rate each of those advertisers on the extent to which the above statements describe them. A four-point scale was used in which 4 means the statement describes them extremely well and 1 means the statement does not describe them at all. By correlating the ads’ statement ratings with their likelihood of being selected, we are able to derive how important those statements are to the decision to call an advertiser. Table 1 shows the ranking of the statements by their correlation with the ranking of the ads. It shows that high quality and trustworthiness are the most important drivers of customers’ decisions to call directory advertisers.
The differences between color or process photo ads and knockout or yellow and black ads are quite pronounced in terms of these characteristics. This can be seen in Figures 5 and 6.
Having shown that color enhances an image of trustworthiness and quality in an advertiser, it remains for us to quantify the impact of these image characteristics on the selection of an ad. For this, we had three of the image statements (trustworthiness, high quality and family orientation) incorporated into the logit model in place of color penetration. Results appear in Figure 7. These were the only attributes included because the others did not have coefficients that were significant at the 95 percent confidence level. These attributes explained about one quarter of the variance in the data.
Color works
Color ads increase calls from prospective customers (which, one might expect, would generate more sales). And not only do they stand out, they also alter the way potential customers look at advertisers. Those who use color are regarded as more trustworthy and as offering higher quality merchandise or services. Thus, color’s impact persists even as greater penetration of color makes individual color ads stand out less. This information will enable directory publishers to position color ads more effectively in order to sell more of them.
The use of an original experimental design and realistic props to collect data, combined with a logit model to analyze the data, were necessary to answer a difficult marketing question. This application of marketing research techniques illustrates the value of research in contributing to the bottom line of corporations who use it.